3 Replies Latest reply on Sep 15, 2018 2:55 PM by Ken Flerlage

# Sankey challenge! One side is sum(measure). The other side is countd(dimension).

I've only done Sankey diagrams in which the stacked bar charts on either side use the same measure. I'm wondering if there's another way, though. I admit this is purely aesthetic, but I wonder if it's possible. Something like this, only with sigmoidal lines instead of straight ones:

I'll attach a sample workbook. I did the bottom bar as countd so that I could get a nice even distribution to match the horizontal container field of customers below.

Let me know if you think it's possible, and if you have any ideas how we might be able to do it!

Thanks,

Vince

• ###### 1. Re: Sankey challenge! One side is sum(measure). The other side is countd(dimension).

How's this?

See attached workbook. Most of the calculations have been borrowed from the sankey template from my website: Chart Templates Part 1: Sankeys - Ken Flerlage: Analytics Architecture, Strategy, & Visualization

Adjustments have been made to show flow between measures related to a single dimension (instead of flow of a single measure from/to separate dimensions).

The chart has two data sources: 1) Superstore data and 2) Model (same model as on the template noted above). These are joined using a cross join (uses a 1=1 join calculation).

See attached data and packaged workbook.

• ###### 2. Re: Sankey challenge! One side is sum(measure). The other side is countd(dimension).

Looks like you nailed it, Ken. I figured this would be one that looked simple at a glance but was quite complicated under the hood. I look forward to hearing how you arrived at the solution though, and seeing whether it's a solution befitting a one-off or a maintainable visualization. Really great job!

• ###### 3. Re: Sankey challenge! One side is sum(measure). The other side is countd(dimension).

Thanks Vince. Under the covers, it's pretty much the same calculations as a sankey, which as you know, are fairly complex. But, with the basic workbook and the model template, it's not too difficult to add in your own data. That being said, if you have a lot of non-aggregated data, then it's going to cross join that to the model data, duplicating each record 98 times. So, ideally, you'd want to use pre-aggregated data.

As discussed, I'll write up a more thorough explanation of all this and put it in a blog. I think it'll probably be a while before I get a chance to do that though, so if you have any specific questions, let me know.

Thanks for the challenge!!